Impacts of widespread badger culling on cattletuberculosis: concluding analyses from alarge-scale field trial

Christl A. Donnelly a,b,*, Gao Wei a, W. Thomas Johnston a, D.R. Cox b,c,Rosie Woodroffe b,d, F. John Bourne b, C.L. Cheeseman e,Richard S. Clifton-Hadley f, George Gettinby b,g, Peter Gilks a,Helen E. Jenkins a, Andrea M. Le Fevre a, John P. McInerney b,W. Ivan Morrison b,h

International Journal of Infectious Diseases (2007) 11, 300—308

http://intl.elsevierhealth.com/journals/ijid

aDepartment of Infectious Disease Epidemiology, Faculty of Medicine, Imperial College London, Norfolk Place,London W2 1PG, UKb Independent Scientific Group on Cattle TB, c/o Defra, London, UKcNuffield College, Oxford, UKdDepartment of Wildlife, Fish & Conservation Biology, University of California, Davis, California, USAeCentral Science Laboratory, Sand Hutton, York, UKfVeterinary Laboratories Agency, Addlestone, Surrey, UKgDepartment of Statistics and Modelling Science, University of Strathclyde, Glasgow, UKhCentre for Tropical Veterinary Medicine, University of Edinburgh, Roslin, Midlothian, UK

Received 3 April 2007; received in revised form 19 April 2007; accepted 20 April 2007Corresponding Editor: William Cameron, Ottawa, Canada

KEYWORDSBovine TB;Mycobacterium bovis;Badger culling;Cattle

Summary

Background: Bovine tuberculosis (TB) has re-emerged as a major problem for British cattlefarmers. Failure to control the infection has been linked to transmission from European badgers;badger culling has therefore formed a component of British TB control policy since 1973.Objectives and design: To investigate the impact of repeated widespread badger culling on cattleTB, the Randomised Badger Culling Trial compared TB incidence in cattle herds in and around tenculling areas (each 100 km2) with those in and around ten matched unculled areas.Results: Overall, cattle TB incidence was 23.2% lower (95% confidence interval (CI) 12.4—32.7%lower) inside culled areas, but 24.5% (95% CI 0.6% lower—56.0% higher) higher on land �2 kmoutside, relative tomatched unculled areas. Inside the culling area boundary the beneficial effectof culling tended to increase with distance from the boundary ( p = 0.085) and to increase onsuccessive annual culls (p = 0.064). In adjoining areas, the detrimental effect tended to diminishon successive annual culls (p = 0.17). On the basis of such linear trends, the estimated net effect

* Corresponding author. Tel.: +44 20 7594 3394.E-mail address: c.donnelly@imperial.ac.uk (C.A. Donnelly).

1201-9712/$32.00 # 2007 International Society for Infectious Diseases. Published by Elsevier Ltd. All rights reserved.doi:10.1016/j.ijid.2007.04.001

Impacts of widespread badger culling on cattle tuberculosis 301

per annum for culling areas similar to those in the trial was detrimental between the first and secondculls, but beneficial after the fourth and later culls, for the range of analyses performed.Conclusions: Careful consideration is needed to determine in what settings systematic repeatedculling might be reliably predicted to be beneficial, and in these cases whether the benefits of suchculling warrant the costs involved.# 2007 International Society for Infectious Diseases. Published by Elsevier Ltd. All rights reserved.

Introduction

Bovine tuberculosis (TB) is a serious disease of cattle, whichhas re-emerged as a major problem for British cattle farmersin the last two decades.1 From the 1950s onwards, routinetesting of cattle combined with slaughter of affected ani-mals, successfully eliminated the infection from much ofBritain. However, infection remained in some areas, and thiswas linked to infection of local badger (Meles meles) popula-tions with Mycobacterium bovis, the causative agent of thedisease.1 To try to reduce transmission of infection frombadgers to cattle, badger culling formed a component ofBritish TB control policy from 1973 until the start of theRandomised Badger Culling Trial (RBCT). Both the incidenceand geographical extent of cattle TB have been increasingsince the 1980s.1

The RBCTwas launched in 1998 to evaluate the effective-ness of badger culling as a control strategy for cattle TB inBritain.2 The RBCT involved comparing the incidence ofcattle TB under three experimental treatments — repeatedwidespread (‘proactive’) culling, localized (‘reactive’) cul-ling, and no culling (‘survey only’) — each replicated tentimes in large (100 km2) trial areas recruited as matched setsof three, known as ‘triplets’. Results from the RBCT, pub-lished shortly after the suspension of reactive culling3,4 andthe completion of proactive culling,5 showed that wide-spread proactive culling reduced cattle TB incidence insidethe culled areas, but elevated incidence in neighboringunculled areas, apparently because culling induced changesin badger behavior, which increased the transmission ofinfection both between badgers, and from badgers to cat-tle.6,7 Localized reactive culling likewise was associated withan overall detrimental effect,3,4 apparently for similar eco-logical reasons.

If culling-induced changes to badger behavior and ecologydo indeed influence TB dynamics, several predictions can bemade about the impact of culling on TB risks for cattle. First,the benefits of proactive culling would be expected toincrease as repeated culls are performed. This is becausesurveys for signs of badger activity indicate that badgerdensity decreased with repeated proactive culling.8

A second prediction is that the beneficial effects of badgerculling should vary with distance inside the culling areaboundary. This is because the proportion of badgers capturedclose to the culling area boundary increased on successiveculls (probably due to immigration from surrounding areas),8

indicating that a more thorough removal was sustained dee-per inside trial areas. Additionally, the initial prevalence ofM. bovis infection in badgers was lower close to culling areaboundaries than deeper inside, but on successive culls pre-valence rose more markedly close to the boundaries.7 If M.bovis prevalence in badgers reflects TB risk to cattle, these

differences would be expected to influence the beneficialeffects of culling.

A third prediction is that the effects of badger cullingcould be influenced by geographical barriers to badger move-ment. The published association between repeated proactiveculling and increased M. bovis infection in badgers wasobserved only in trial areas where landscape conditionsallowed badgers to immigrate into culled areas from neigh-boring land; no such effect was seen where coastline, majorrivers or motorways formed a substantial proportion of trialarea boundaries.7 Hence, geographical barriers to badgermovement might also affect the impact of culling on cattleTB.

A fourth and final prediction is that the beneficial effectsof badger culling might be influenced by land access. Land-holder consent was required before field staff could survey orcull badger populations. Every trial area contained landwhere consent was refused, and land for which no landholdercould be identified. No traps were set on such land, althoughefforts were made to capture badgers residing in these areasby trapping around their boundaries. Nevertheless, the ben-efits of proactive culling observed on accessible land mightbe expected to be greater than those observed overall.Additionally, since cattle TB incidence was elevated onunculled land immediately outside culling areas, similardetrimental effects might be expected on land inside trialareas that was not accessible for culling.

All of these predictions have implications for the design ofany future badger culling strategies. Here, we extend pre-vious analyses5 of the impacts of proactive culling usingfurther data, and investigate whether such impacts variedwith successive culls, herd location relative to trial areaboundaries, permeability of trial area boundaries for bad-gers, and land accessibility for culling. We also explore theimpact of land access on badger removal rates.

Materials and methods

Trial design and implementation

Thirty trial areas, each about 100 km2, were recruited as tenmatched triplets, located in areas of high cattle TB incidence(Figure 1). In most cases trial area boundaries followed prop-erty boundaries, butwhere possible, they followedgeographicfeatures likely to impede badger movement (coastline, rivers,large conurbations, motorways and dual carriageways).Boundary permeability was measured as the proportion ofthe perimeter not composed of such geographical barriers.7

Neighboring trial areaswere separated by buffer zones at least3 kmwide.All trial areaswere surveyed forbadger activity andthen randomly allocated to treatments (except in one tripletwhere security concerns directed a specific allocation) such

302 C.A. Donnelly et al.

Figure 1 Locations of RBCT trial areas, denoting areas allo-cated to proactive culling (solid shading), reactive culling(hatched shading), and survey-only (open). Parish testing inter-val gives an indication of local cattle TB incidence. (Reproducedfrom the supplementary information to Donnelly CA, WoodroffeR, Cox DR, Bourne FJ, Cheeseman CL, Wei G, et al. Positive andnegative effects of widespread badger culling on cattle tuber-culosis. Nature 2006;439:843—6.).

that each treatment — proactive culling, reactive culling, orsurvey only (no culling) —was repeated ten times, once withineach triplet.

Initial consent was sought from landholders before areaswere surveyed for badger activity and culling treatmentsallocated. Landholders consented to surveying and culling(‘culling’ land), surveying but not culling (‘survey’ land), orrefused all access (‘refusal’ land). Additionally, each trialarea contained land for which no landholder could be iden-tified (‘unsigned’ land). Landholders could change consentstatus at any time. Following treatment allocation, initialculls were conducted on all land in the proactive areas forwhich consent was given. Culling treatment area boundarieswere defined (beyond trial area boundaries where necessary)using field survey data to ensure that all badgers likely to useland inside the trial areas were targeted. Badgers werecaptured in cage traps placed primarily at setts. Trappingwas suspended in February—April each year to avoid killingmothers with dependent cubs still confined to the sett.9,10

Few badgers sustained trap-related injuries,11,12 and dis-patch (by gunshot) was deemed ‘humane’ by independentaudit.13—17 Initial culls for each proactive trial area werecompleted between December 1998 and December 2002, and‘follow-up’ culls were repeated approximately annually(with longer delays incurred due to a nationwide epidemicof foot-and-mouth disease in 2001).7

As soon as the initial proactive cull was complete, datawere collected on cattle TB incidence in and around trial

areas, using established veterinary surveillance. Surveillancedistinguished ‘confirmed breakdowns’ (incidents in whichpostmortem examination of slaughtered cattle led to detec-tion of TB lesions or culture of M. bovis) from ‘unconfirmedbreakdowns’ (incidents in which one or more cattle reactedto the tuberculin test but infection was not confirmed atpostmortem or by culture). Here our primary analyses werebased on the incidence of confirmed breakdowns; analyses ofall breakdowns (confirmed and unconfirmed), and uncon-firmed breakdowns only, are presented and discussed insupplementary material.

Analyses of the effect of land accessibility used the ear-liest available data on the consent status of each land parcel,since interpretation of consent-stratified analyses of cattleTB incidence would be invalid if changes in consent dependedon culling-induced changes in the local risk of cattle TB(although no evidence of such dependence has been found,see supplementary material for details). The proportion ofinaccessible (‘survey’, ‘refusal’ or ‘unsigned’) land withinproactive treatment areas varied from 15% to 50% (30%overall). In total, 73% of inaccessible land was within200 m of accessible (‘culling’) land (see supplementarymaterial).

Effects of proactive culling on TB incidence incattle

Statistical analysis methods were similar to those previouslypublished.5 Log-linear Poisson regression was used to com-pare the numbers of breakdowns recorded in trial areassubjected to the proactive and survey-only treatments,adjusting for triplet, the log of the number of baseline herdsat risk, and the log of the number of confirmed breakdownsrecorded either one or three years prior to RBCT culling. Tominimize bias in the covariates, 0.5 was added to the num-bers of baseline herds and historic confirmed breakdownsbefore the logs were taken.18 Other measures of the at-riskpopulation were explored in supplementary material. As inprevious analyses,5 confidence intervals (CI) and p-valueswere conservatively adjusted for extra-Poisson overdisper-sion by using an adjustment factor (the square root of themodel deviance divided by the degrees of freedom) in allcases where its value was greater than 1.

Cattle herd locations were taken from two alternativedatabases, the national animal health information systemVetNet, and a separate database set up specifically for theRBCT. Analyses performed using these two databases arepresented separately. These databases were used to identifyherds inside, and up to 2 km outside, trial area boundaries.The VetNet database provided more complete data on herdsoutside trial areas, because the RBCT database did notinclude all farms on neighboring land. Herds within 2 km ofmore than one trial area boundary (whether proactive,reactive or survey-only) were omitted from analyses.

Primary analyses covered the period from the initialproactive cull in each triplet, to a date one year after cullinghad ceased in that triplet, when another cull would haveoccurred had proactive culling continued. This time period —which totalled 55.8 triplet-years — also offered an opportu-nity for annual herd testing to detect any breakdowns thatoccurred during the culling period. Estimates obtained in thecurrent analyses may differ slightly from previously published

Impacts of widespread badger culling on cattle tuberculosis 303

estimates for the same time periods due to an improvementin the linking of herds to their breakdowns in the surveillancedatabase. Additional estimates of the effect of proactiveculling were obtained from stratified data for accessible andinaccessible land, for different time periods, and for differ-ent distances from the trial area boundary. Tests for lineartrends (on the log scale) with the number of repeat culls anddistance from the trial area boundary were performed usingweighted least squares. The effects of boundary permeabilitywere investigated by adding a main effect of permeabilityand a treatment � permeability interaction term to the log-linear models, tests of the latter being of interest. Severalother factors (e.g., measures of initial badger density) werehypothesized to influence the impact of culling; these aredetailed in supplementary material. The findings based onspatial—temporal variation within trial areas are indepen-dent of the randomization used in the primary comparisons ofthe trial.

Effects of land access on proactive badgerremoval

Badgers were trapped only on land inside treatment areaswhere ‘cull’ consent had been given. However, efforts weremade to capture badgers resident on inaccessible land byplacing traps on nearby accessible land. To assess the effec-tiveness of these efforts we analyzed capture rates within200 m of inaccessible land within trial areas (the 200 mdistance being dictated by the precision of recording capturelocations on 1:10 000 maps and the need to place traps atlocations such as badger setts or paths). To avoid includingbadgers immigrating from outside the proactive treatmentarea, we excluded land <200 m inside the treatment areaboundary. Parallel analyses investigated whether findings

Table 1 Estimated effects of proactive culling on the incidence ofbaseline herds, and historic TB incidence (over three years)

Proactive effect

Estimate 95%

Inside trial areasUsing VetNet location dataFrom initial cull (cull 1) �23.2% (�3From first follow-up cull (cull 2) �26.6% (�3Between initial and follow-up �7.2% (�3

Using RBCT location dataFrom initial cull (cull 1) �17.4% (�2From first follow-up cull (cull 2) �21.0% (�3Between initial and follow-up 1.1% (�2

Up to 2 km outside trial areasUsing VetNet location dataFrom initial cull (cull 1) 24.5% (�0From first follow-up cull (cull 2) 19.6% (�1Between initial and follow-up 46.8% (�0

Using RBCT location dataFrom initial cull (cull 1) 35.3% (5.8From first follow-up cull (cull 2) 24.9% (�7Between initial and follow-up 95.4% (10

RBCT, Randomised Badger Culling Trial.

were influenced by badger preference for any propertyboundary (irrespective of consent status) but gave similarresults (see supplementary material).

The badger removal rate on accessible land (less the200 m zones around inaccessible land) was compared withthat in the 200 m zones, andwith that in the inaccessible landand 200 m zones combined. Numbers of badgers caught werecompared using log-linear models, adjusting for triplet andlog-transformed land area (km2). We also investigated inter-actions between badger capture locations and cull type(initial vs. follow-up).

Results

During the entire study period there were 472 confirmedbreakdowns among herds identified by VetNet as being insidethe proactive trial areas, and 362 confirmed breakdowns onland up to 2 km outside. Inside proactive trial areas, 245herds had a single confirmed breakdown, 74 herds had twoconfirmed breakdowns, and 25 herds hadmore than two (3.16on average). On the land up to 2 km outside proactive trialareas, 191 herds had a single confirmed breakdown, 58 herdshad two confirmed breakdowns, and 17 herds had more thantwo (3.24 on average).

Overall effects of proactive culling on TBincidence in cattle

The primary analysis revealed that the overall incidence ofconfirmed TB breakdowns in cattle was 23.2% (95% CI 12.4—32.7%) lower inside proactively culled trial areas than insideunculled survey-only areas ( p < 0.001). This effect was con-sistent across all ten proactive/survey-only pairs (the test foroverdispersion was not significant, p = 0.87). As in previous

confirmed cattle TB breakdowns. Analyses adjusted for triplet,

Overdispersion

CI p-Value Factor p-Value

2.7%, �12.4%) <0.001 0.67 0.876.8%, �14.8%) <0.001 0.93 0.531.3%, 25.4%) 0.63 1.05 0.36

7.2%, �6.2%) 0.003 0.79 0.741.6%, �8.8%) 0.001 0.86 0.646.4%, 39.0%) 0.94 1.15 0.23

.6%, 56.0%) 0.057 1.26 0.130.3%, 59.5%) 0.22 1.41 0.052.4%, 116.4%) 0.052 0.95 0.50

%, 73.0%) 0.016 1.00 0.44.2%, 67.9%) 0.14 1.09 0.34.5%, 245.5%) 0.021 0.82 0.69

304 C.A. Donnelly et al.

Figure 2 (A) Variation in the effect of proactive culling by thenumber of repeat culls. (B) Variation in the effect of proactiveculling at different distances from the trial area boundary. Theseanalyses used cattle herd locations from the VetNet databaseand adjusted for historic cattle TB incidence (over three years).Error bars denote 95% confidence intervals.

analyses,5 this effect was somewhat stronger whenmeasuredfrom the first follow-up cull (cull 2), rather than the initialcull (cull 1; Table 1). Similar results were achieved if theRBCT database (rather than the VetNet database) was used toidentify cattle herds inside trial areas (Table 1), and ifanalyses adjusted for historic incidence in the previous year(rather than the previous three years; details in supplemen-tary material). Furthermore, findings were robust to the useof different measures of the size of the cattle population atrisk (see supplementary material).

On land up to 2 km outside proactive trial areas, overallcattle TB incidence was 24.5% higher (95% CI 0.6% lower—56.0% higher) than that on land neighboring survey-only areas( p = 0.057). Once again, the effect was consistent across allten proactive/survey-only pairs (the test for overdispersionwas not significant, p = 0.13). Similar patterns were detectedusing herd locations from the RBCT database, and adjustingfor one year’s historic incidence (see supplementary mate-rial). Interestingly, this detrimental effect of culling wasmost marked between the initial and first follow-up cull;weaker detrimental effects were detected after the firstfollow-up cull (Table 1).

These results can be used to estimate the number ofconfirmed breakdowns prevented, and induced, by proactiveculling. If we assume that a 100-km2 circular area wereculled, then just under 83.5 km2 of land is within 2 km ofthe culling area boundary. If the herd density were1.25 per km2 (roughly that seen in trial areas), then therewill be 125 herds in the culling area and 104 in the neighbor-ing area. If the underlying incidence rate throughout ten suchareas were 8 confirmed breakdowns per 100 herds per year,then these results are equivalent to the saving of an esti-mated 116 fewer confirmed breakdowns (10 areas � 125herds � 8 confirmed breakdowns/100 herds per year � 5years � 0.232) over 5 years in the culling areas and anestimated 102 additional breakdowns (10 areas � 104herds � 8 confirmed breakdowns/100 herds per year � 5years � 0.245) over 5 years in the neighboring areas due toproactive culling, for a net benefit of 14 fewer breakdownsover 5 years across the ten 183.5 km2 combined areas. If theunderlying incidence rate were lower in the neighboring areathan in the culling area, then this net benefit would begreater. However, unless this underlying rate is considerablylower than that in the culling area, the 95% predictioninterval for the net benefit will include zero.

Effect of repeated proactive culling on cattle TBincidence

When the data were stratified based on the intervals betweensuccessive culls (initial to second, second to third, third tofourth, and after fourth), the beneficial effect of proactiveculling inside trial areas appeared to increase with repeatedculling (Figure 2A). The linear trend (on the log scale)suggested an 11.2% increase in the beneficial proactive effectwith each cull ( p = 0.064). Correspondingly, the detrimentaleffect of proactive culling in the neighboring area appearedto diminish with each cull (Figure 2A), with the linear trendsuggesting a 7.3% decrease in the detrimental proactiveeffect with each cull ( p = 0.17). Based on the estimates fromthe models with linear trends, the estimated net effect perannum appeared detrimental between the first and second

culls, but beneficial after the third and later culls. Similarnon-significant trends were found adjusting for one year’shistoric incidence and using herd locations from the RBCTdatabase adjusting for three years’ historic incidence (seesupplementary material). For the range of analyses per-formed, the estimated net effect per annumwas detrimentalbetween the first and second culls, but beneficial after thefourth and later culls.

Dependence of the effects of proactive culling onproximity to the boundary

The beneficial effect of proactive culling appeared toincrease at greater distances inside the trial area boundary(Figure 2B, p = 0.085). Detrimental effects of culling wereobserved for herds 0.5—2 km outside the trial area boundary,while those within 0.5 km of the trial area boundaryappeared to experience a benefit (Figure 2B). This lattereffect was unsurprising, because badger culling extendedbeyond the boundaries of the trial areas to target socialgroups judged, on the basis of field signs, to occupy homeranges falling partially inside the trial areas.

Impacts of widespread badger culling on cattle tuberculosis 305

There was no evidence that this dependence in the effectof culling on proximity to the boundary changed in responseto repeated culling (see supplementary material).

Effects of boundary permeability

In badgers, the permeability of culling area boundaries waspreviously shown to have a marked effect on how M. bovisprevalence changed on successive culls.6 However, the over-all effect of culling on cattle TB was not significantly modifiedby boundary permeability either inside trial areas ( p = 0.73)or on neighboring lands ( p = 0.69). As only two proactive trialareas had more than 20% of their boundaries relativelyimpermeable to badgers, it was not possible to stratifyanalyses of time trends by boundary permeability.

Effects of land access on the impact of proactiveculling on cattle TB

Comparing TB incidence in herds on accessible proactive landwith entire survey-only areas indicated effects of cullingcomparable in magnitude and precision with those observedon proactive land as a whole (Table 2). Effects of culling oninaccessible proactive land were estimated less precisely andwere thus less consistent between models using the VetNetand RBCT databases. As the estimates from inaccessible landwere of limited precision, it was unsurprising that compar-isons between effect estimates based on accessible andinaccessible land showed no significant differences( p = 0.36 using herd locations from the VetNet database,Table 2).

Effects of land access on badger removal rates

Data on capture rates suggest that substantial numbers ofbadgers were removed from inaccessible land by trapping inthe surrounding 200 m zones. If badger density was uniformacross trial areas and no badgers were taken from inacces-sible land then, based on the relative area of accessible land,inaccessible land, and 200 m zones, 44% fewer captures

Table 2 Estimated effect of proactive culling on the incidencebaseline herds, and historic TB incidence (over three years). All herdonly trial areas, and then the comparison is stratified by consent

Sourceof herdlocationdata

Consent status Baseline herds(proactive/survey-only)

Proactive effect

Estimate 95% CI

VetNet All proactive landb 1221/1276 �23.2% (�32.7Accessible 812/817 �15.4% (�29.9Inaccessible 409/459 �28.7% (�48.6

RBCT All proactive landb 1009/1127 �17.4% (�27.2Accessible 834/873 �15.5% (�28.1Inaccessible 175/254 �10.6% (�42.4

RBCT, Randomised Badger Culling Trial.a 95% CI for the difference in the effect of proactive badger culling be

71.0%; RBCT, �41.0%, 51.7%.b Estimates comparable to those presented in Donnelly et al.,5 but usi

that triplet.

would be expected per km2 in 200 m zones plus inaccessibleland than on the remaining accessible land. In fact, on initialculls, only 28% fewer badgers were taken from each km2 of200 m zones plus inaccessible land, with the 95% CI (8%—43%fewer, p = 0.007 for the hypothesis of no difference betweencapture rates) indicating a removal rate significantly greaterthan expected (44% fewer, p = 0.033). This effect differed(interaction p = 0.014) between initial and follow-up culls; onfollow-ups there was no difference in removal rate betweenland types with different accessibility (3% fewer on inacces-sible land plus 200 m zones; 95% CI 17% more—20% fewer;p = 0.74).

On initial culls, there was no difference between badgercapture rates in the 200 m zones and on remaining accessibleland (2% more per km2 in 200 m zones; 95% CI 18% fewer—26%more, p = 0.88). On follow-up culls more badgers may havebeen culled per km2 in the 200 m zones (18% more, 95% CI 1%fewer—39% more, p = 0.060). However, the differencebetween removal rates in 200 m zones and the remainingaccessible land did not vary significantly between initial andfollow-up culls ( p = 0.29).

Discussion

The results presented here extend those published pre-viously, and provide further evidence that the impact ofbadger culling on cattle TB incidence depends critically uponthe geographic extent and temporal pattern of culling. Thisinformation will be very important in deciding whetherbadger culling should contribute to future strategies to con-trol cattle TB.

The inclusion of additional data confirmed our previousfinding — that proactive culling has the capacity to bothdecrease and increase the incidence of cattle TB5 — butimproved the precision of the estimates. As predicted, thebeneficial effects of proactive culling tended to becomemore marked on successive culls; this is consistent withthe finding that badger density inside proactive areasdeclined over the same period.8 By contrast, the detrimentaleffects of culling detected on neighboring land were initially

of confirmed TB breakdowns. Analyses adjusted for triplet,s in proactive trial areas are compared with all those in survey-status of land

Overdispersion p-Value fordifferencebetween accessibleand inaccessible ap-Value Factor p-Value

%, �12.4%) <0.001 0.67 0.87 —%, 2.0%) 0.080 1.33 0.009 0.36%, �1.0%) 0.044

%, �6.2%) 0.003 0.79 0.74 —%, �0.6%) 0.042 1.25 0.034 0.82%, 38.6%) 0.615

tween herds on accessible and inaccessible land: VetNet, �17.8%,

ng incidence data up to a date one year after culling had ceased in

306 C.A. Donnelly et al.

substantial, but may have become less strong on later culls.As a result, the estimated net impact of proactive cullingobserved in the RBCT transitioned from being detrimentalbetween the initial and second culls, to being beneficial afterthe fourth and later culls.

The balance of benefit and detriment would clearly beaffected by the size of the area culled (and hence the relativeproportions of land falling inside, and immediately outside,the culling area). Likewise, our analyses illustrate the sensi-tivity of the outcome to the baseline incidence of cattle TBinside and outside the culling areas. In the RBCT, trial areaswere centered on areas of high cattle TB risk, so baselineincidence was higher in the culling (and matched survey-only) areas than on neighboring land. Had trial areas beenlocated such that background incidence was equivalentinside and immediately outside the areas culled (but withthe same underlying proportions of breakdowns due to bad-gers inside and immediately outside the areas as in the trial),the estimated absolute number of breakdowns inducedwould have been greater, reducing the overall benefits.

Our findings confirm that infectious contact betweenbadgers and cattle is related to badger density in a mannerthat is strongly non-linear. At the comparatively high popula-tion densities that were observed inside survey-only areas,and would have occurred in proactive areas prior to culling,badgers occupy small territories and M. bovis infection ishighly clustered.6,19—22 This territorial organization presum-ably limits, to some extent, the risk of cattle herds encoun-tering infectious badgers. Substantially lowering density byrepeated proactive culling disrupted territorial organization,expanded movement patterns, and elevated M. bovis pre-valence in badgers,6,7 but presumably reduced badger—cat-tle contact to such an extent that incidence was reduced incattle. This reduction was not achieved immediately, prob-ably because more thorough badger removal was sustainedon later culls.8

A different pattern occurred on unculled land immedi-ately outside proactive culling areas. Badger densities werereduced only slightly, but spatial organization was disruptedand movement patterns were expanded, as in culled areas.6

Because expanded ranging was observed in the absence of amajor reduction in density, the risks of infectious contactbetween badgers, and between badgers and cattle, wereprobably elevated, rather than reduced, by culling. Thereduction in density was smallest immediately after initialculls8 and this pattern, together with field evidence thatbadgers alter their movement patters rapidly in response todisturbance,23—25 may explain why the detrimental effect oncattle TB was particularly marked at this time (Figure 2A).

We speculate that the reduced detrimental effectobserved in neighboring areas on later culls (Figure 2A)may have reflected changes in the structure of badger popu-lations prompted by repeated culling. Although social dis-ruption has been shown to alter badger movement patternson a timescale of days or weeks,23—25 depletion of the badgerpopulation on land neighboring trial areas may have takenseveral years. This is because culling is likely to haveprompted not only expansion of badgers’ daily ranges (asmeasured in the RBCT6) but also increased dispersal (that is,a permanent shift of an individual’s home range from onelocation to another; this occurs more commonly in badgerpopulations living at low population densities25). Dispersal

can take several months to complete26 and, in young animalsat least, is likely to be seasonal; hence dispersal into culledareas may continue to occur in the months or years aftercompletion of each cull. However, sincemost dispersal occursover short distances (1—2 territories in undisturbed popula-tions25,27), repeated culling is likely to have eventuallydepleted the pool of dispersers and allowed a quasi-stablespatial organization to re-emerge. This might explain thereduced detrimental effect apparently recorded on neigh-boring lands on later proactive culls. Unfortunately, thishypothesis cannot be tested because monitoring of badgeractivity across successive culls was confined to the trial areasthemselves, and not conducted outside. However, if thisinterpretation were correct, it would suggest that an equiva-lent reduction in the detrimental effect would probably notbe achieved by conducting the same number of culls over amuch shorter period. There was no evidence that the dis-tribution of detrimental effects, relative to the culling areaboundary, changed consistently over time (see supplemen-tary material).

As expected, the beneficial effect of badger culling oncattle TB appeared more marked at greater distances insidethe trial area boundary (Figure 2B). This was predicted bothbecause badger removal was more thorough deeper insidetrial areas,8 and also because the prevalence of M. bovisinfection in badgers was higher on such land, particularly oninitial culls.7 However, as there was no evidence that thedependence of the beneficial effects on proximity to the trialarea boundary changed over successive culls (see supplemen-tary material), this pattern is most likely to reflect thedegree to which badger population densities were reducedat greater distances from the culling area boundary.

The effect of culling onM. bovis prevalence in badgers wasinfluenced by the permeability of trial area boundaries.7

However, no such effect on the incidence of cattle TB wasdetected, either inside or outside trial area boundaries. Thismay be because of limited statistical power: the RBCTwas notdesigned to test this hypothesis2 and the variation amongtrial areas in boundary permeability was not great.7 Thus,currently available data shed no direct light on whether aproactive culling policy would be more beneficial — or lessdetrimental — if conducted in more geographically isolatedareas.

The prevalence of M. bovis infection in badgers was alsoelevated following a nationwide epidemic of foot-and-mouthdisease, during which routine cattle TB testing was sus-pended.7 However, there was no evidence that this influ-enced the beneficial or detrimental effects of badger cullingon cattle TB (details in supplementary material).

There was no evidence that lack of consent to cull (at thelevel and configuration observed in the RBCT) underminedthe beneficial effects of badger culling: the incidence ofcattle TB on accessible land was comparable with thatobserved overall and not significantly different from thaton inaccessible land. This contrasts with the situationobserved immediately outside trial areas, where herds onunculled land experienced elevated TB incidence. Thisdifference probably reflects the substantial ecological dif-ferences between the two types of unculled land. Inacces-sible areas were mostly small relative to badger home rangesize: 95% of the inaccessible land fell within 500 m of cullingareas, comparable with the median ranging distances of

Impacts of widespread badger culling on cattle tuberculosis 307

badgers in proactive areas (300—700 m).6 As most inacces-sible land was completely surrounded by culled land, therewere opportunities to capture badgers resident on inacces-sible land in traps placed on nearby accessible land. Incontrast, survey data indicate that the unculled land sur-rounding proactive treatment areas contained large, con-tiguous and mostly intact badger populations, which couldprovide a source of colonists to re-occupy land cleared byculling. Hence, the dynamics of badger—badger and bad-ger—cattle interactions would be expected to differ in thetwo types of unculled land.

These findings have important implications for decisionsabout whether badger culling should contribute to futurestrategies to control cattle TB.28 Results confirm that badgerculling is only likely to be beneficial if conducted system-atically over large areas, and sustained over several years.Previous analyses have suggested that culling conducted insmall, localized areas has overall detrimental effects,3,4 andanalyses presented here indicate that the detrimentaleffects of widespread culling outweigh the benefits if cullingis not sustained. Our results also suggest that careful deploy-ment of traps around inaccessible land contributed to thebeneficial effects of culling reported here. Detailed consid-eration would be needed to determine whether this level ofeffort would be feasible in the long term, and whether thelevel of disease control achieved would be sufficient towarrant the costs involved.

Acknowledgements

This study was funded and implemented by the Departmentof Environment, Food and Rural Affairs (Defra). We gratefullyacknowledge the contribution made by the staff of Defra andits associated agencies. We also wish to thank the manyfarmers and land occupiers in the trial areas who allowedthe experimental treatments to operate on their land.

Conflict of interest statement: The members (FJB, CAD,DRC, GG, JPM, WIM and RW) of the Independent ScientificGroup on Cattle TB (ISG) receive fees for time spent on ISGwork and reimbursement of expenses from the Departmentof Environment, Food and Rural Affairs (Defra).

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.ijid.2007.04.001.

References

1. Krebs JR, Anderson R, Clutton-Brock T, Morrison I, Young D,Donnelly C, et al. Bovine tuberculosis in cattle and badgers.London: HMSO; 1997.

2. Bourne J, Donnelly C, Cox D, Gettinby G, McInerney JP, MorrisonI, et al. Towards a sustainable policy to control TB in cattle–—ascientific initiative. First report of the Independent ScientificGroup on Cattle TB. London: Ministry of Agriculture, Fisheries &Food; 1998.

3. Donnelly CA, Woodroffe R, Cox DR, Bourne J, Gettinby G, LeFevre AM, et al. Impact of localized badger culling on TBincidence in British cattle. Nature 2003;426:834—7.

4. Le Fevre AM, Donnelly CA, Cox DR, Bourne FJ, Clifton-Hadley RS,Gettinby G, et al. The impact of localised reactive badgerculling versus no culling on TB incidence in British cattle: arandomised trial. London: Department for Environment, Foodand Rural Affairs; 2005. Available at: http://www.defra.gov.uk/animalh/tb/isg/pdf/lefevre1005.pdf (accessed April 2007).

5. Donnelly CA, Woodroffe R, Cox DR, Bourne FJ, Cheeseman CL,Wei G, et al. Positive and negative effects of widespread badgerculling on cattle tuberculosis. Nature 2006;439:843—6.

6. Woodroffe R, Donnelly CA, Cox DR, Bourne FJ, Cheeseman CL,Delahay RJ, et al. Effects of culling on badger (Meles meles)spatial organization: implications for the control of bovinetuberculosis. J Appl Ecol 2006;43:1—10.

7. Woodroffe R, Donnelly CA, Jenkins HE, Johnston WT, Cox DR,Bourne FJ, et al. Culling and cattle controls influence tuberculosisrisk for badgers. Proc Natl Acad Sci U S A 2006;103:14713—7.

8. Woodroffe R, Gilks P, JohnstonWT, Le Fevre AM, Cox DR, DonnellyCA, et al. Effects of culling on badger abundance: implicationsfor tuberculosis control. J Zool (accepted for publication).

9. Woodroffe R, Bourne FJ, Cheeseman CL, Cox DR, Donnelly CA,Gettinby G, et al. Welfare of badgers (Meles meles) subjected toculling: development and evaluation of a closed season. AnimWelfare 2005;14:19—25.

10. Woodroffe R, Donnelly CA, Wei G, Bourne FJ, Cheeseman CL, CoxDR, et al. Evaluating the closed season adopted in the Rando-mised Badger Culling Trial–—an update. London: Department forEnvironment, Food and Rural Affairs; 2007. Available at: http://www.defra.gov.uk/animalh/tb/isg/pdf/closedseason.pdf(accessed April 2007).

11. Woodroffe R, Bourne FJ, Cheeseman CL, Cox DR, Donnelly CA,Gettinby G, et al. Welfare of badgers (Meles meles) subjected toculling: patterns of trap-related injury. Anim Welfare2005;14:11—7.

12. Woodroffe R, Donnelly CA, Wei G, Bourne FJ, Cheeseman CL, CoxDR, et al. Patterns of trap-related injury recorded in theRandomised Badger Culling Trial–—an update. London: Depart-ment for Environment, Food and Rural Affairs; 2007. Availableat: http://www.defra.gov.uk/animalh/tb/isg/pdf/trapinjur-ies.pdf (accessed April 2007).

13. Kirkwood JK. Humaneness of MAFF’s badger despatch proce-dures. London: Ministry of Agriculture, Fisheries and Food; 2000.Available at: http://www.defra.gov.uk/animalh/tb/publica-tions/auditor/report.htm (accessed April 2007).

14. Ewbank R. Report of the second independent auditor on thehumaneness of dispatch procedures used in the randomisedbadger culling trial and Defra’s response. London: Departmentfor Environment, Food and Rural Affairs; 2003 . Available at:http://www.defra.gov.uk/animalh/tb/publications/pdf/rep_indaud.pdf (accessed April 2007).

15. Ewbank R. The third independent audit on the humaneness ofdispatch procedures used in the randomised badger culling trialand Defra’s response. London: Department for Environment,Food and Rural Affairs; 2004. Available at: http://www.defra.-gov.uk/animalh/tb/pdf/humrep2004.pdf (accessed April 2007).

16. Anderson J. The fourth independent audit on the humaneness ofdispatch procedures used in the randomised badger culling trialand Defra’s response. London: Department for Environment,Food and Rural Affairs; 2005. Available at: http://www.defra.-gov.uk/animalh/tb/culling/humane2005audit.pdf (accessedApril 2007).

17. Anderson J. The fifth independent audit on the humaneness ofdispatch procedures used in the randomised badger culling trialand Defra’s response. London: Department for Environment,Food and Rural Affairs; 2006. Available at: http://www.defra.-gov.uk/animalh/tb/publications/pdf/humaneness-audit06.pdf(accessed April 2007).

18. Cox DR. Some statistical methods connected with series ofevents. J Roy Stat Soc B 1955;17:129—64.

308 C.A. Donnelly et al.

19. Woodroffe R, Donnelly CA, Johnston WT, Bourne FJ, CheesemanCL, Clifton-Hadley RS, et al. Spatial association of Mycobacter-ium bovis infection in cattle and badgers. J Appl Ecol2005;42:852—62.

20. Cheeseman CL, Jones GW, Gallagher J, Mallinson PJ. The popu-lation structure, density and prevalence of tuberculosis (Myco-bacterium bovis) in badgers (Meles meles) from four areas insouth-west England. J Appl Ecol 1981;18:795—804.

21. Cheeseman CL, Little TW, Mallinson PJ, Page RJ, Wilesmith JW,PritchardDG.Populationecologyandprevalenceof tuberculosis inbadgers inanareaofStaffordshire.MammalRev1985;15:125—35.

22. Delahay RJ, Langton S, Smith GC, Clifton-Hadley RS, CheesemanCL. The spatio-temporal distribution of Mycobacterium bovis(bovine tuberculosis) infection in a high-density badger popula-tion. J Anim Ecol 2000;69:428—41.

23. Cheeseman CL, Mallinson PJ, Ryan J, Wilesmith JW. Recolonisa-tion by badgers in Gloucestershire. In: Hayden TJ, editor. Thebadger. Dublin: Royal Irish Academy; 1993. p. 78—93.

24. Roper TJ, Lups P. Disruption of territorial behaviour in badgersMeles meles. Z Saugetierkd 1993;58:252—5.

25. Woodroffe R, Macdonald DW, da Silva J. Dispersal and philo-patry in the European badger, Meles meles. J Zool 1995;237:227—39.

26. Roper TJ, Ostler JR, Conradt L. The process of dispersal inbadgers (Meles meles). Mammal Rev 2003;33:314—8.

27. Rogers LM, Delahay R, Cheeseman CL, Langton S, Smith GC,Clifton-Hadley RS. Movement of badgers (Meles meles) in a high-density population: individual, population and disease effects.Proc R Soc Lond B 1998;265:1269—76.

28. Department for Environment, Food and Rural Affairs. Controllingthe spread of bovine tuberculosis in cattle in high incidenceareas in England: badger culling. A consultation document issuedby the Department for Environment, Food and Rural Affairs.London: Department for Environment, Food and Rural Affairs;2005. Available at: http://www.defra.gov.uk/animalh/tb/pdf/badger-consult.pdf (accessed May 2007).